Don't touch that moonshine!
It'll make you go blind!!
Then again I don't care, because when it comes to AI sorry - Super High Intelligence Technology, i'm teetotal.
Irony alert: OpenAI whines that Chinese model stole its special IP that it stole from everybody else
OpenAI, which hoovered up vast amounts of internet content amid copyright fights, has accused individuals associated with China’s Moonshot AI of being involved in a "distillation attack" that began July 1. The house of Altman warns that extracting its models’ reasoning at scale could help rivals train capable models without …
-
-
Thursday 1st October 2026 09:23 GMT steviebuk
Much like facebook. That were using pirated books to train their model. They knew the books were pirated as they were using a torrent to grab them. And they were actively trying to do this away from their network to hide that it was facebook torrenting. And put their seeding so low so they could claim, they weren't really sharing that torrent that much.
How do we know this? Because its all in the court documents. What was the outcome, when it was an open and shut case, with clear evidence they were using pirated books? Any normal person like you or me, bang to rights, prison time, massive fine.
Well nope, the judge said "Oh its fine. They are allowed". FFS. If no brown envelopes were thrown around then, then maybe facecloth have compromising photos on said judge.
-
-
-
Thursday 1st October 2026 13:15 GMT doublelayer
Re: really?
I think it's mostly an attempt to ban their competition. If it's illegal to pay for non-US LLMs and they can also prevent open weights versions of the same models being used, then they're the only ones left. It might not work, but what's the harm in throwing another argument at regulators who have so far bought almost every argument about how AI is brilliant and that we must remove everything that might pose the slightest difficulty to our heroic LLM companies mastering this new, crucial enterprise?
-
-
Thursday 1st October 2026 13:18 GMT doublelayer
Re: really?
I don't think it is illegal, other than a terms of service fight which is the LLM companies' problem, not mine or any governments'. I expect many of the queries are very cheap for the distillers since a lot of companies are still offering subscriptions which allow people to chat way more than the subscription itself costs. They've been forcing up prices for people using harnesses or which they guess might be doing so, but text chat, all you need to run distillation queries, is often still subsidized.
-
Thursday 1st October 2026 17:32 GMT Sproggit
Depends on the "Illegal" in question...
I'm not trying to be a smart-alec, I'm trying to tease out the difference between a breach-of-contract and criminal law.
We need to look at both of those endpoints to determine the exact nature of what has happened here...
At the Criminal end of that spectrum we've got the Computer Fraud and Abuse Act, 18 U.S. Code § 1030. That would come in to play if the activities deliberately circumvented e.g. access controls or other safeguard that OpenAI applied to their models to ensure all access was legitimate and authorized.
At the Civil end of that spectrum we've got basic Contract and Tort law, since all user's of OpenAI's services are required to accept OpenAI's terms of service, which are rendered under US law and which include e.g provisions such as arbitration clauses that tip the scales massively in OpenAI's favor.
From a read of the article and the way that Altman suggests the exfiltration was done, it looks as though this could not be described as a 1030 violation [at least there is no claim as such]; rather this is described as a violation of OpenAI's terms of service. Typically, we'd refer to a breach of criminal law as "illegal" and a contract breach being a "civil dispute". A civil dispute can also become illegal, of course, but typically only if the actions of a party to the dispute can be shown to be so.
Still a reasonable grounds for complaint, but very much more in the "good luck with that" corner, rather than having the full weight of the President's personal Law Firm - sorry, the Department of Justice - stamping down on things.
And just like the US government has been quite happy to tell others [like Australia, when their Medicare web site got hacked by OpenAI] to "go pound sand"... this is likely an instance where the US is going to have to do that instead.
-
-
-
Thursday 1st October 2026 09:00 GMT Bebu sa Ware
“Adversarial distillation poses safety and national security risks,”
And non-adversarial distillation is ?
"Preserved thinking" — pickled brains ?
The whole AI (SI) circus was always a clown show but further descended into a theatre of the ridiculous that would have challenged even Jarry to replicate, but suppose the same might equally be said of the US as a whole.
-
Thursday 1st October 2026 13:05 GMT heyrick
Dear China
The bot knows about me, my blog, my writing style, etc. So it has clearly ingested my site as part of its training.
I own the rights on the crap that I wrote, and not OpenAI, therefore I hereby grant you permission to make use of any of my site/data that OpenAI have used (in particular without having the decently to discuss with me prior to ripping me off to enrich themselves (and fuck up the memory and storage markets as a side quest)). Knock yourselves out...
-
